WEBVTT 00:00:00.000 --> 00:00:02.419 Imagine just blinking your eyes to take a photo 00:00:02.419 --> 00:00:05.379 or waking up to find an invisible assistant has 00:00:05.379 --> 00:00:08.359 already prepped your morning meetings. The AI 00:00:08.359 --> 00:00:11.779 future isn't a chat box anymore. It's becoming 00:00:11.779 --> 00:00:14.919 a continuous layer over reality itself. Yeah, 00:00:14.939 --> 00:00:17.160 that chat window we all got used to. It's basically 00:00:17.160 --> 00:00:20.199 dead. The technology is completely dissolving 00:00:20.199 --> 00:00:23.179 into the background of everything we do. Welcome 00:00:23.179 --> 00:00:25.379 to the deep dive. Google just dropped a tidal 00:00:25.379 --> 00:00:27.859 wave of announcements at IO 2026. We've got a 00:00:27.859 --> 00:00:29.920 massive stack of updates to synthesize for you 00:00:29.920 --> 00:00:32.700 today. Right. The central theme here is invisible 00:00:32.700 --> 00:00:36.020 workflow level integration. We're tracking how 00:00:36.020 --> 00:00:38.579 AI is physically escaping the browser. We're 00:00:38.579 --> 00:00:40.640 going to start with the underlying engine, Gemini 00:00:40.640 --> 00:00:43.380 Omni. Then we'll explore the total redesign of 00:00:43.380 --> 00:00:46.119 search. And we'll get into world building creative 00:00:46.119 --> 00:00:49.219 tools plus Android XR glasses and the efficient 00:00:49.219 --> 00:00:51.820 new model powering it all. Let's jump right into 00:00:51.820 --> 00:00:54.479 how this new AI brain actually processes our 00:00:54.479 --> 00:00:56.320 physical reality. Well, the biggest paradigm 00:00:56.320 --> 00:00:58.700 shift with Gemini Omni is moving away from text 00:00:58.700 --> 00:01:02.399 -first thinking. It uses native multimodal processing. 00:01:02.539 --> 00:01:05.439 Which means processing text, audio, images, and 00:01:05.439 --> 00:01:08.760 video all at the exact same time. Exactly. It 00:01:08.760 --> 00:01:10.980 doesn't separate them anymore. Wait, I'm stuck 00:01:10.980 --> 00:01:13.439 on something right out of the gate. If Omni is 00:01:13.439 --> 00:01:16.040 analyzing all that simultaneously, how is it 00:01:16.040 --> 00:01:18.879 not just melting the servers? Because it's not 00:01:18.879 --> 00:01:21.549 translating between them anymore. Old AI video 00:01:21.549 --> 00:01:24.209 generation was like a really messy assembly line. 00:01:24.349 --> 00:01:26.590 You'd write a text script. Right. Then hand it 00:01:26.590 --> 00:01:28.950 to an audio model. Then another completely separate 00:01:28.950 --> 00:01:31.629 model tried guessing the visual layout. It was 00:01:31.629 --> 00:01:34.250 horribly inefficient. You lose so much context 00:01:34.250 --> 00:01:36.530 passing data down the line like that. Yeah, you 00:01:36.530 --> 00:01:39.469 do. Omni acts much more like a master architect. 00:01:39.590 --> 00:01:42.150 You give it one prompt and it conceptualizes 00:01:42.150 --> 00:01:44.709 the entire house simultaneously. The script, 00:01:44.950 --> 00:01:47.250 the visual physics, the acoustics of the room. 00:01:47.370 --> 00:01:49.349 It's all just one native thought. But the part 00:01:49.349 --> 00:01:52.849 that really caught my eye was its grasp of real 00:01:52.849 --> 00:01:54.950 -world physics. It actually understands that 00:01:54.950 --> 00:01:58.090 an apple falls down, not up. Gravity is mathematically 00:01:58.090 --> 00:02:01.170 baked into its core logic, which completely changes 00:02:01.170 --> 00:02:03.549 how it generates visual media. Think about the 00:02:03.549 --> 00:02:05.989 incredible demo they showed. The prompt asking 00:02:05.989 --> 00:02:08.330 for a watercolor animation of a seed growing. 00:02:08.490 --> 00:02:10.550 Yeah, and it had to be explained simply for a 00:02:10.550 --> 00:02:12.830 10 -year -old. A year ago, that prompt would 00:02:12.830 --> 00:02:15.789 give you a glitchy, terrifying mess. Right, and 00:02:15.789 --> 00:02:17.909 the tree branches would randomly morph into human 00:02:17.909 --> 00:02:21.460 fingers. But Omni creates a scientifically accurate 00:02:21.460 --> 00:02:24.639 fluid animation. The roots grow downward through 00:02:24.639 --> 00:02:27.740 the dirt correctly. The water absorption follows 00:02:27.740 --> 00:02:31.099 actual biological rules and it dynamically adds 00:02:31.099 --> 00:02:34.340 a friendly fully voiced explanation layered perfectly 00:02:34.340 --> 00:02:37.479 over the visuals. So how exactly does baking 00:02:37.479 --> 00:02:41.439 in this world understanding prevent those weird 00:02:41.439 --> 00:02:43.659 hallucinatory videos we're so used to seeing? 00:02:43.939 --> 00:02:46.930 It relies on a physical grounding model. Basically, 00:02:47.090 --> 00:02:50.090 the AI constantly checks its visual output against 00:02:50.090 --> 00:02:52.389 the actual laws of physics. Like calculating 00:02:52.389 --> 00:02:55.129 mass and light. Exactly. It calculates how mass, 00:02:55.189 --> 00:02:57.610 light, and gravity actually behave in 3D space. 00:02:58.069 --> 00:03:00.009 If it tries to draw a shadow pointing toward 00:03:00.009 --> 00:03:02.590 the sun, the physics engine flags it as impossible. 00:03:02.949 --> 00:03:05.030 It autocorrects the generated pixel before you 00:03:05.030 --> 00:03:07.840 ever see it. So applying real physics keeps the 00:03:07.840 --> 00:03:10.539 AI from generating impossible dreamlike visual 00:03:10.539 --> 00:03:14.139 mistakes. Beat. Yeah. It firmly anchors the digital 00:03:14.139 --> 00:03:16.939 imagination to our physical reality. Now, if 00:03:16.939 --> 00:03:19.560 the AI understands our physical world that well, 00:03:19.759 --> 00:03:21.580 let's look at how it navigates the digital one. 00:03:21.580 --> 00:03:23.439 We have to talk about Google search. Oh, this 00:03:23.439 --> 00:03:26.120 is genuinely the biggest redesign of search in 00:03:26.120 --> 00:03:29.780 25 years. The classic list of blue links is entirely 00:03:29.780 --> 00:03:32.560 gone. It functions much more like an AI operating 00:03:32.560 --> 00:03:35.439 system now. It builds dynamic visual. cards on 00:03:35.439 --> 00:03:38.120 the fly? You aren't opening 10 separate tabs 00:03:38.120 --> 00:03:41.219 to compare things anymore. The interface physically 00:03:41.219 --> 00:03:44.719 adapts to your specific problem. It uses these 00:03:44.719 --> 00:03:48.060 new AI search agents to handle complex multi 00:03:48.060 --> 00:03:50.580 -step logic. It does. It totally shifts the workload. 00:03:50.819 --> 00:03:53.219 Give me a practical example of that. What happens 00:03:53.219 --> 00:03:55.259 if I want to find a new hobby? Let's say you're 00:03:55.259 --> 00:03:57.500 searching for local pottery classes. You don't 00:03:57.500 --> 00:04:00.300 just type the keyword anymore. You write a complex 00:04:00.300 --> 00:04:03.039 query. Like asking it to compare prices and check 00:04:03.039 --> 00:04:05.879 weekend availability. Exactly. Compare prices, 00:04:06.219 --> 00:04:08.340 check schedules, and explain the difference between 00:04:08.340 --> 00:04:11.319 their throwing techniques. The agent acts like 00:04:11.319 --> 00:04:13.319 an investigative researcher. That's a lot of 00:04:13.319 --> 00:04:15.740 separate web searches. But the AI pulls data 00:04:15.740 --> 00:04:18.680 from all those local sites instantly. It builds 00:04:18.680 --> 00:04:21.600 one custom comparison dashboard just for you. 00:04:21.819 --> 00:04:24.480 It does all the heavy lifting. Which brings us 00:04:24.480 --> 00:04:27.209 to Gemini Spark and the Daily Brief. If search 00:04:27.209 --> 00:04:29.970 organizes the web, these organize your actual 00:04:29.970 --> 00:04:32.589 life. They're absolute game changers for saving 00:04:32.589 --> 00:04:36.069 your daily mental bandwidth. Gemini Spark safely 00:04:36.069 --> 00:04:38.750 reads your workspace apps. It connects your Gmail, 00:04:39.050 --> 00:04:42.149 Google Docs, and Calendar into one cohesive brain. 00:04:42.610 --> 00:04:44.470 Imagine you have a high -stakes client meeting 00:04:44.470 --> 00:04:48.810 tomorrow. Spark quietly scans your inbox. It 00:04:48.810 --> 00:04:51.769 finds the three background PDFs your boss forwarded 00:04:51.769 --> 00:04:54.069 last week. Right. It reads them, writes a half 00:04:54.069 --> 00:04:56.660 -page summary, and attaches it directly to the 00:04:56.660 --> 00:04:58.920 calendar invite. You wake up fully prepped. And 00:04:58.920 --> 00:05:00.639 then you have the daily brief. You open your 00:05:00.639 --> 00:05:02.779 phone in the morning, and it aggregates everything 00:05:02.779 --> 00:05:06.139 perfectly. It highlights urgent emails from VIPs. 00:05:06.279 --> 00:05:08.800 It warns you about heavy rain during your commute. 00:05:08.980 --> 00:05:11.759 It even flags overlapping appointments. It strips 00:05:11.759 --> 00:05:13.620 away all the digital noise. You don't have to 00:05:13.620 --> 00:05:15.439 triage five different apps before you've even 00:05:15.439 --> 00:05:17.720 had your coffee. I do have to push back a little 00:05:17.720 --> 00:05:20.519 here, though. If search just hands us the perfectly 00:05:20.519 --> 00:05:23.920 curated answer, do we lose the serendipity of 00:05:23.920 --> 00:05:26.540 discovering things ourselves? We used to stumble 00:05:26.540 --> 00:05:28.819 onto weird, fascinating websites by accident. 00:05:29.000 --> 00:05:32.279 It's a really valid concern. You do lose that 00:05:32.279 --> 00:05:34.699 random website wandering. But think about what 00:05:34.699 --> 00:05:38.399 you actually gain. You bypass the SEO optimized 00:05:38.399 --> 00:05:41.860 fluff completely. You get immense focus. The 00:05:41.860 --> 00:05:44.199 serendipity shifts to what you actively build. 00:05:44.420 --> 00:05:46.920 Exactly. The serendipity shifts from what you 00:05:46.920 --> 00:05:49.300 passively find online to what you build with 00:05:49.300 --> 00:05:51.779 the three hours you just saved. We trade endless 00:05:51.779 --> 00:05:54.360 scrolling for instant multi -step problem solving 00:05:54.360 --> 00:05:57.439 right in the search bar. Beat? It treats your 00:05:57.439 --> 00:05:59.759 attention as your absolute most valuable asset. 00:06:00.019 --> 00:06:02.139 We've seen what it finds for us, but the interface 00:06:02.139 --> 00:06:04.399 itself, how we actually talk to it, is getting 00:06:04.399 --> 00:06:06.060 completely out of the way. Google is calling 00:06:06.060 --> 00:06:08.779 this their neural expressive design. It uses 00:06:08.779 --> 00:06:11.959 bright colors and fluid glowing animations. It's 00:06:11.959 --> 00:06:14.220 designed to feel alive and highly responsive 00:06:14.220 --> 00:06:16.509 when it's thinking. The Gemini desktop app is 00:06:16.509 --> 00:06:18.709 the perfect example of this. It's totally screen 00:06:18.709 --> 00:06:21.689 aware. It watches your open Chrome tabs and listens 00:06:21.689 --> 00:06:25.170 to you while you work. And I have a vulnerable 00:06:25.170 --> 00:06:27.410 admission to make here. I still wrestle with 00:06:27.410 --> 00:06:30.050 prompt drift myself. Losing track of your original 00:06:30.050 --> 00:06:31.910 thought halfway through a voice command. Exactly. 00:06:32.069 --> 00:06:36.029 I start asking the AI to format a document. Then 00:06:36.029 --> 00:06:39.550 I remember I need an email drafted and I just 00:06:39.550 --> 00:06:42.230 stumble over my words completely. I end up rambling 00:06:42.230 --> 00:06:45.269 into the microphone. Almost everyone does. We 00:06:45.269 --> 00:06:47.910 don't naturally speak in perfectly structured 00:06:47.910 --> 00:06:50.329 written paragraphs. We really don't. That's why 00:06:50.329 --> 00:06:52.370 the new desktop app is so clever. You can just 00:06:52.370 --> 00:06:55.569 speak your messy chaotic thoughts out loud. The 00:06:55.569 --> 00:06:57.790 app listens patiently through your long pauses. 00:06:58.230 --> 00:07:00.889 It ignores your ums and ahs. It extracts the 00:07:00.889 --> 00:07:02.829 actual intent from the rambling. Yeah, it cleans 00:07:02.829 --> 00:07:06.329 up the audio and instantly outputs a sharp professional 00:07:06.329 --> 00:07:08.410 text draft of what you actually meant to say. 00:07:08.550 --> 00:07:10.810 It's like having a hyper -competent secretary 00:07:10.810 --> 00:07:13.509 translating your brain fog. We're seeing this 00:07:13.509 --> 00:07:16.009 frictionless design on mobile, too, with Gemini 00:07:16.009 --> 00:07:19.149 Live. It uses a dynamic island on your phone 00:07:19.149 --> 00:07:22.209 screen. The voice mode opens seamlessly as a 00:07:22.209 --> 00:07:24.930 small overlay. It sits right on top of whatever 00:07:24.930 --> 00:07:26.550 web page you're currently reading. You don't 00:07:26.550 --> 00:07:28.370 get kicked out to a separate chat screen. And 00:07:28.370 --> 00:07:30.769 you can interrupt the AI effortlessly. Right. 00:07:30.910 --> 00:07:32.670 If it's summarizing an article and going too 00:07:32.670 --> 00:07:34.449 slow, you just say stop and ask a new question. 00:07:35.170 --> 00:07:38.550 The animation creates this gentle glowing wave 00:07:38.550 --> 00:07:41.810 effect when you speak. It naturally mirrors the 00:07:41.810 --> 00:07:45.589 cadence of a real human conversation. But I do 00:07:45.589 --> 00:07:48.180 have to ask about privacy here. What are the 00:07:48.180 --> 00:07:51.379 actual privacy implications of an AI constantly 00:07:51.379 --> 00:07:54.160 watching your active screen? That sounds a little 00:07:54.160 --> 00:07:57.399 dystopian. It relies on a very strict, explicit 00:07:57.399 --> 00:08:01.000 boundary. The desktop app is not recording your 00:08:01.000 --> 00:08:03.319 entire operating system in the background. Okay, 00:08:03.319 --> 00:08:05.680 so what does it see? It only ever sees the data 00:08:05.680 --> 00:08:08.759 you explicitly highlight or the specific window 00:08:08.759 --> 00:08:11.339 you drag into its view. It only analyzes the 00:08:11.339 --> 00:08:13.199 specific window you actively choose to share 00:08:13.199 --> 00:08:16.180 with it. Right. You control the exact perimeter 00:08:16.180 --> 00:08:17.930 of its awareness. We're going to take a quick 00:08:17.930 --> 00:08:21.589 break. Stick around. MidRule sponsor, ReadPlaceholder. 00:08:21.889 --> 00:08:24.269 And we are back. Welcome back to the Deep Dive. 00:08:24.730 --> 00:08:27.269 We just talked about how AI understands our intent 00:08:27.269 --> 00:08:29.930 and our screen. Because it has that context, 00:08:30.250 --> 00:08:32.570 we can now use it to build entirely new tools 00:08:32.570 --> 00:08:35.629 without knowing a single line of code. The traditional 00:08:35.629 --> 00:08:37.870 technical barrier to entry has essentially vanished 00:08:37.870 --> 00:08:39.789 overnight. We're talking about antigravity 2 00:08:39.789 --> 00:08:42.690 .0. This is their new AI coding environment. 00:08:42.940 --> 00:08:44.860 If you're a software engineer listening right 00:08:44.860 --> 00:08:48.600 now, hearing the phrase, teams of AI agents building 00:08:48.600 --> 00:08:50.600 software, probably makes your blood pressure 00:08:50.600 --> 00:08:53.139 spike. It sounds like an automatic job replacement. 00:08:53.580 --> 00:08:55.399 But this isn't about replacing the architect, 00:08:55.559 --> 00:08:57.480 it's about replacing the bricklayer. You don't 00:08:57.480 --> 00:08:59.799 stare at complicated lines of syntax anymore. 00:09:00.039 --> 00:09:02.860 No. You act as the manager for a team of specialized 00:09:02.860 --> 00:09:06.120 AI agents. You give them natural language instructions, 00:09:06.299 --> 00:09:08.480 and they construct the app. Let's unpack the 00:09:08.480 --> 00:09:10.659 specific creative tools they rolled out, starting 00:09:10.659 --> 00:09:12.919 with Flow. This lets you build custom editing 00:09:12.919 --> 00:09:15.419 tools just by describing them. Say you're editing 00:09:15.419 --> 00:09:18.000 a video and you repeatedly need to isolate the 00:09:18.000 --> 00:09:20.679 sky and boost the contrast. Instead of doing 00:09:20.679 --> 00:09:23.860 that manually every time, you type, make sky 00:09:23.860 --> 00:09:27.200 bright blue. Flow instantly codes a custom button 00:09:27.200 --> 00:09:29.799 into your workspace that does exactly that. It 00:09:29.799 --> 00:09:32.279 turns repetitive manual labor into a single click. 00:09:32.899 --> 00:09:35.379 Then there's Stitch, which is aimed at web design. 00:09:35.450 --> 00:09:38.389 It integrates directly into your existing Figma 00:09:38.389 --> 00:09:40.950 files. The most impressive part of Stitch is 00:09:40.950 --> 00:09:45.669 the native AI micro -edits. Usually, if an AI 00:09:45.669 --> 00:09:48.610 tries to redesign a web page, it breaks the entire 00:09:48.610 --> 00:09:51.230 layout. It hallucinates a new structure. It messes 00:09:51.230 --> 00:09:54.759 up the CSS completely. Exactly. But Stitch safely 00:09:54.759 --> 00:09:57.480 isolates specific visual elements. You can tell 00:09:57.480 --> 00:09:59.740 it to redesign just one specific button. And 00:09:59.740 --> 00:10:02.440 it alters that local code without cascading errors 00:10:02.440 --> 00:10:04.639 through the rest of your page layout. Yeah. That 00:10:04.639 --> 00:10:07.159 is incredible. So if Stitch handles the underlying 00:10:07.159 --> 00:10:09.460 web page structure, what happens when the actual 00:10:09.460 --> 00:10:11.759 images on that page need to change? That brings 00:10:11.759 --> 00:10:15.259 us to Google Pix. Pix is their clean, AI -first 00:10:15.259 --> 00:10:18.000 photo editor. It focuses heavily on object -level 00:10:18.000 --> 00:10:20.080 editing. You can grab a coffee cup in a photo, 00:10:20.299 --> 00:10:22.820 drag it across a table, and the AI perfectly 00:10:22.820 --> 00:10:24.799 autofills the background where the cup used to 00:10:24.799 --> 00:10:27.419 be. It even handles editable text while preserving 00:10:27.419 --> 00:10:30.179 the original shadows. And for audio, they introduced 00:10:30.179 --> 00:10:33.039 flow music. You get granular control over specific 00:10:33.039 --> 00:10:34.919 instruments. You don't have to re -record an 00:10:34.919 --> 00:10:37.220 entire track if the bass is slightly off. You 00:10:37.220 --> 00:10:39.840 just ask the AI to tweak the bassline genre. 00:10:40.059 --> 00:10:43.019 Plus, it hooks into Omni to generate perfectly 00:10:43.019 --> 00:10:45.919 synced music videos. Let me ask the obvious developer 00:10:45.919 --> 00:10:48.720 question here. How does anti -gravity prevent 00:10:48.720 --> 00:10:51.899 bad, hallucinated code from cascading into a 00:10:51.899 --> 00:10:54.399 completely broken application? It uses continuous 00:10:54.399 --> 00:10:57.659 automated testing. These AI agents don't just 00:10:57.659 --> 00:11:00.580 write code, they deploy it in safe sandboxes. 00:11:00.659 --> 00:11:03.360 They detect their own runtime errors. Yes, and 00:11:03.360 --> 00:11:05.659 they rewrite the logic until the software functions 00:11:05.659 --> 00:11:08.460 perfectly. The AI agents constantly test and 00:11:08.460 --> 00:11:10.399 self -correct their own code without any human 00:11:10.399 --> 00:11:13.539 intervention. Yeah, it builds the software and 00:11:13.539 --> 00:11:15.580 heals the software at the exact same time. So 00:11:15.580 --> 00:11:17.340 far, we've talked about manipulating digital 00:11:17.340 --> 00:11:20.139 screens, but Google is moving aggressively to 00:11:20.139 --> 00:11:23.419 project the digital world directly into our physical 00:11:23.419 --> 00:11:25.940 3D space. This is where the announcements felt 00:11:25.940 --> 00:11:29.159 truly futuristic. Let's dive into Project Genie. 00:11:29.539 --> 00:11:32.639 This is a model dedicated to generating interactive, 00:11:33.100 --> 00:11:37.320 fully playable 3D worlds to sex silence. Whoa. 00:11:38.160 --> 00:11:41.679 Imagine generating explorable dynamic 3D environments 00:11:41.679 --> 00:11:45.700 from 20 years of Maps Street View data. It is 00:11:45.700 --> 00:11:47.399 staggering to think about. You aren't just looking 00:11:47.399 --> 00:11:50.080 at flat photos anymore. You can simulate dynamic 00:11:50.080 --> 00:11:52.360 real -world environments. You could literally 00:11:52.360 --> 00:11:55.179 walk down a digital recreation of a Parisian 00:11:55.179 --> 00:11:57.659 street from 2008. It generates the geometry, 00:11:57.879 --> 00:12:00.360 the textures, the lighting, all of it. Which 00:12:00.360 --> 00:12:02.940 ties perfectly into their hardware push. They're 00:12:02.940 --> 00:12:05.340 officially entering the display glasses market 00:12:05.340 --> 00:12:08.230 with Android XR. The wearable tech community 00:12:08.230 --> 00:12:10.429 has been holding its breath for this. They split 00:12:10.429 --> 00:12:12.529 it into two distinct categories. Right, first 00:12:12.529 --> 00:12:14.769 you have the audio glasses. These look exactly 00:12:14.769 --> 00:12:16.730 like regular fashion frames, complete normal 00:12:16.730 --> 00:12:18.769 sunglasses. But they have directional speakers 00:12:18.769 --> 00:12:21.090 and a microphone to converse with the Gemini 00:12:21.090 --> 00:12:23.450 Assistant. Then you have the heavy hitters, the 00:12:23.450 --> 00:12:26.049 display glasses. These actually project augmented 00:12:26.049 --> 00:12:28.470 reality holograms directly into your field of 00:12:28.470 --> 00:12:31.360 vision. The practical features are wild. It provides 00:12:31.360 --> 00:12:34.340 AR navigation where glowing directional arrows 00:12:34.340 --> 00:12:36.460 are painted directly onto the sidewalk in front 00:12:36.460 --> 00:12:39.100 of you. And the camera integration is brilliant. 00:12:39.600 --> 00:12:41.899 You literally just bling your eyes twice to snap 00:12:41.899 --> 00:12:44.240 a photo of whatever you're looking at. It entirely 00:12:44.240 --> 00:12:46.659 removes the phone from the equation. But how 00:12:46.659 --> 00:12:49.539 do you actually interact with pop -up notifications 00:12:49.539 --> 00:12:51.840 if you can't use your hands to tap a screen? 00:12:52.120 --> 00:12:54.740 They built a really elegant spatial interface. 00:12:55.129 --> 00:12:58.590 When a text comes in, the alert doesn't blindly 00:12:58.590 --> 00:13:01.429 block your physical view of the world. It floats 00:13:01.429 --> 00:13:03.870 gently in your peripheral vision. You just speak 00:13:03.870 --> 00:13:06.429 a quick command like read it or dismiss and it 00:13:06.429 --> 00:13:09.029 reacts. Exactly. Notifications gently appear 00:13:09.029 --> 00:13:11.350 in your vision and you manage them entirely using 00:13:11.350 --> 00:13:15.309 your voice. Beat. Right. The goal is to keep 00:13:15.309 --> 00:13:17.990 you present in the real world while staying connected. 00:13:18.200 --> 00:13:20.960 When you have technology this immersive and intelligent, 00:13:21.159 --> 00:13:23.200 we have to talk about the real -world stakes. 00:13:23.620 --> 00:13:26.000 This isn't just about fun consumer gadgets anymore. 00:13:26.159 --> 00:13:29.580 Not at all. It's actively solving massive professional 00:13:29.580 --> 00:13:32.120 and scientific bottlenecks right now. Look at 00:13:32.120 --> 00:13:34.480 small business branding with a tool like Pomelli. 00:13:34.860 --> 00:13:37.600 You feed it a rough, half -baked business idea. 00:13:37.759 --> 00:13:40.259 And it instantly builds out a cohesive brand 00:13:40.259 --> 00:13:43.419 identity, logo, and marketing copy. Or look at 00:13:43.419 --> 00:13:45.700 the medical field. where Gemini is being deployed 00:13:45.700 --> 00:13:48.340 to read thousands of dense medical research papers 00:13:48.340 --> 00:13:50.820 simultaneously. It cross -references millions 00:13:50.820 --> 00:13:53.860 of data points to identify potential new medical 00:13:53.860 --> 00:13:56.700 treatments. It spots complex biochemical patterns 00:13:56.700 --> 00:13:59.539 that human researchers simply don't have the 00:13:59.539 --> 00:14:02.139 time or cognitive capacity to see. The scale 00:14:02.139 --> 00:14:04.740 of pattern recognition is unparalleled. We're 00:14:04.740 --> 00:14:06.799 seeing the same thing with global weather forecasting. 00:14:07.080 --> 00:14:10.919 The AI analyzes massive data sets of ocean currents 00:14:10.919 --> 00:14:13.279 and atmospheric pressure. It predicts the path 00:14:13.279 --> 00:14:16.779 of major catastrophic storms days faster than 00:14:16.779 --> 00:14:19.200 traditional supercomputers. It literally gives 00:14:19.200 --> 00:14:22.000 coastal cities more time to evacuate. That directly 00:14:22.000 --> 00:14:24.799 saves human lives. But as the capability scales, 00:14:25.019 --> 00:14:27.679 so does the risk, especially with media generation. 00:14:28.000 --> 00:14:30.399 We're dealing with AI that generates audio and 00:14:30.399 --> 00:14:33.179 video that looks and sounds completely indistinguishable 00:14:33.179 --> 00:14:36.169 from reality. which brings us to the most critical 00:14:36.169 --> 00:14:49.240 security announcement, Synthely. Break down how 00:14:49.240 --> 00:14:51.740 that actually works. If someone generates a fake 00:14:51.740 --> 00:14:53.980 emit, can't they just crop it or slap a filter 00:14:53.980 --> 00:14:56.860 on it to strip the synthide watermark off? No, 00:14:56.860 --> 00:14:59.100 because it's not a stamp sitting on top of the 00:14:59.100 --> 00:15:01.820 photo. Think of it like mixing blue dye into 00:15:01.820 --> 00:15:04.320 a glass of water. Once it's stirred in, you can 00:15:04.320 --> 00:15:06.759 pour half the water down the drain, which is 00:15:06.759 --> 00:15:08.919 like cropping the photo. You can even freeze 00:15:08.919 --> 00:15:11.360 it. But the remaining water is still fundamentally 00:15:11.360 --> 00:15:13.980 blue. The watermark is baked into the underlying 00:15:13.980 --> 00:15:16.720 data. Right. It is mathematically distributed 00:15:16.720 --> 00:15:20.080 across the entire file. Even if you heavily compress 00:15:20.080 --> 00:15:22.799 or edit the image, specialized verification tools 00:15:22.799 --> 00:15:25.120 can still read the underlying signature. It's 00:15:25.120 --> 00:15:27.740 essential for verifying truth online. The watermark 00:15:27.740 --> 00:15:30.279 is permanently woven into the image pixels, making 00:15:30.279 --> 00:15:33.840 it impossible to remove. Yeah, it survives almost 00:15:33.840 --> 00:15:36.480 any modification you throw at it. None of these 00:15:36.480 --> 00:15:38.799 massive features mean anything if they take five 00:15:38.799 --> 00:15:41.539 minutes to load or if they cost a fortune to 00:15:41.539 --> 00:15:44.639 run. So how is Google making this computationally 00:15:44.639 --> 00:15:47.340 feasible for the average user? Everything we've 00:15:47.340 --> 00:15:49.779 talked about relies on the specific engine underneath. 00:15:50.860 --> 00:15:54.710 They announced Gemini 3 .5 Flash. It's the core 00:15:54.710 --> 00:15:57.389 brainpowering this entire ecosystem. But the 00:15:57.389 --> 00:16:00.009 focus here wasn't on making it a super genius. 00:16:00.509 --> 00:16:04.409 The focus was on extreme efficiency. It is remarkably 00:16:04.409 --> 00:16:07.049 fast and incredibly cheap for Google's servers 00:16:07.049 --> 00:16:09.269 to run. We spent the last two years obsessing 00:16:09.269 --> 00:16:12.350 over godlike AI models that can write symphonies 00:16:12.350 --> 00:16:14.389 or solve quantum physics. But Google realized 00:16:14.389 --> 00:16:16.850 the actual trillion dollar market isn't Mozart. 00:16:17.090 --> 00:16:19.330 The market is a hyper competent digital intern 00:16:19.330 --> 00:16:22.409 that never sleeps. Exactly. Everyday tasks don't 00:16:22.409 --> 00:16:24.789 require massive super intelligence. You don't 00:16:24.789 --> 00:16:26.889 need a digital genius to summarize a Tuesday 00:16:26.889 --> 00:16:28.789 morning calendar invite. You just need it done 00:16:28.789 --> 00:16:32.149 instantly, reliably, and cheaply. And that efficiency 00:16:32.149 --> 00:16:34.370 is what allows Google to scale this globally. 00:16:34.870 --> 00:16:37.210 Their vision is providing a highly capable smart 00:16:37.210 --> 00:16:39.970 assistant to billions of people without charging 00:16:39.970 --> 00:16:42.549 an expensive monthly subscription just for basic 00:16:42.549 --> 00:16:45.360 access. Accessibility is the real moat here. 00:16:45.500 --> 00:16:47.600 They're trying to make intelligence ubiquitous. 00:16:48.240 --> 00:16:51.279 So why prioritize a fast and cheap model over 00:16:51.279 --> 00:16:54.559 a slow genius level model for daily use? Well, 00:16:54.559 --> 00:16:56.620 speed and cost are the only things that let you 00:16:56.620 --> 00:16:58.799 integrate it into actual daily habits. Speed 00:16:58.799 --> 00:17:01.120 and cost are what actually allow them to put 00:17:01.120 --> 00:17:06.019 AI into everyday workflows. Beat? Mm -hmm. If 00:17:06.019 --> 00:17:08.380 it's not instant and invisible, normal people 00:17:08.380 --> 00:17:10.259 just won't use it. So what does this all mean 00:17:10.259 --> 00:17:13.650 for us? If we bring this all home. we're witnessing 00:17:13.650 --> 00:17:17.089 a fundamental paradigm shift. The era of logging 00:17:17.089 --> 00:17:20.710 into a standalone AI chatbot website is officially 00:17:20.710 --> 00:17:24.049 ending. The real value of modern AI is that it's 00:17:24.049 --> 00:17:26.609 becoming an invisible foundational layer. It's 00:17:26.609 --> 00:17:28.470 woven directly into the glasses on our face, 00:17:28.650 --> 00:17:31.009 the workspaces we type in, and the operating 00:17:31.009 --> 00:17:33.250 systems we use to navigate the web. It's just 00:17:33.250 --> 00:17:35.390 becoming part of the fabric of reality. We want 00:17:35.390 --> 00:17:37.190 you to go out and test these boundaries today. 00:17:37.440 --> 00:17:39.880 Throw a highly complex multi -step problem at 00:17:39.880 --> 00:17:42.000 search. See if it can actually build a dashboard 00:17:42.000 --> 00:17:44.059 that changes how you work. But it leaves us with 00:17:44.059 --> 00:17:46.480 something deeper to chew on. What's that? If 00:17:46.480 --> 00:17:48.680 AI successfully removes all the friction from 00:17:48.680 --> 00:17:51.380 our daily tasks, from writing complex software 00:17:51.380 --> 00:17:53.700 code down to summarizing our morning emails, 00:17:54.039 --> 00:17:56.259 what do we actually do with all the quiet space 00:17:56.259 --> 00:17:58.259 it leaves behind? Does removing the friction 00:17:58.259 --> 00:18:00.380 of daily work also remove the spark of human 00:18:00.380 --> 00:18:02.460 struggle that gives it meaning? Think about that 00:18:02.460 --> 00:18:04.759 next time you blink to take a photo. Thanks for 00:18:04.759 --> 00:18:07.079 joining us on this Deep Dive for UTRO Music.